GPU-based database operators using the graphics pipeline
The present disclosure relates to methods and systems for using the computer graphics pipeline to execute database query operations on a graphics processing unit (GPU). The methods and systems use the graphics pipeline to transform relational data into images of the relational data. The methods and systems use the graphics pipeline to perform relational database operations on the images in response to a query and output a query result for the database operations.
1 . A method, comprising:
receiving a query with a relational database operation to perform on relational data;
using a graphics pipeline to perform the relational database operation on the relational data; and
outputting a query result of the relational database operation, wherein the graphics pipeline performs:
a build phase on the relational data to create a hash table of the relational data, wherein the hash table maps the relational data to a stack of two-dimensional (2D) images;
using a flag array to track which attribute values of the relational data are included in the stack of 2D images and which attribute values of the relational data needs processing;
continuing to perform the build phase until the flag array indicates that all the relational data is processed; and
a probe phase that performs the relational database operation by rendering the relational database operation on the hash table and outputting the query result.
2 . The method of claim 1 , wherein the graphics pipeline is executed by a graphics processing unit (GPU) and the graphics pipeline includes a vertex shader, vertex post processing, a fragment shader, and fragment processing.
3 . The method of claim 1 , wherein the relational database operation is an equi-join operation between a first input table of relational data and a second input table of relational data.
4 . The method of claim 3 , wherein the stack of 2D images is created for the first input table of relational data and each 2D image in the stack of 2D images is created in response to a hash collision for the relational data or duplicate attribute values of the relational data.
5 . The method of claim 4 , wherein the probe phase:
performs a lookup for each record in the second input table to each 2D image in the stack of 2D images to identify matches between the relational data in the second input table and attribute values of the relational data in the first input table;
converts the matches to pixels; and
performs post processing that applies color to the pixels indicating matching join records between the relational data in the first input table and the relational data in the second input table and the query result includes the pixels.
6 . The method of claim 1 , further comprising:
performing a count phase that determines an amount of graphics processing unit (GPU) memory required for the relational database operation; and
allocating the amount of GPU memory for the query result.
7 . The method of claim 1 , wherein the relational database operation is a group by operation on an input table of the relational data and each 2D image of the stack of 2D images is created in response to a hash collision for the relational data.
8 . The method of claim 7 , wherein the probe phase:
uses the stack of 2D images to map the relational data to positions defined by a grouping value;
generates pixels with the positions and the grouping values;
performs post processing that applies color to the pixels indicating an aggregate function and attribute values of the relational data; and
performs a blend function of the pixels based on the aggregation function and the query output includes the blended pixels.
9 . The method of claim 8 , wherein the aggregate function includes a count, a sum, a minimum, a maximum, or an average.
10 . The method of claim 1 , wherein the query result is provided to a central processing unit (CPU).
11 . A device, comprising:
a memory to store data and instructions; and
a graphics processing unit operable to communicate with the memory, wherein the graphics processing unit is operable to:
receive a query with a relational database operation to perform on relational data, wherein the relational database operation is an equi-join operation between a first input table of relational data and a second input table of relational data;
use a graphics pipeline to perform the relational database operation on the relational data by performing:
a build phase on the relational data to create a hash table of the relational data, wherein the hash table maps the relational data to a stack of two-dimensional (2D) images and the stack of 2D images is created for the first input table of relational data and each 2D image in the stack of 2D images is created in response to a hash collision for the relational data or duplicate attribute values of the relational data; and
a probe phase that performs the relational database operation by rendering the relational database operation on the hash table; and
output a query result of the relational database operation.
12 . The device of claim 11 , wherein the probe phase:
performs a lookup for each record in the second input table to each 2D image in the stack of 2D images to identify matches between the relational data in the second input table and attribute values of the relational data in the first input table;
converts the matches to pixels; and
performs post processing that applies color to the pixels indicating matching join records between the relational data in the first input table and the relational data in the second input table and the query result includes the pixels.
13 . The device of claim 11 , wherein the graphics pipeline includes a vertex shader, vertex post processing, a fragment shader, and fragment processing.
14 . The device of claim 11 , wherein the query result is provided to a central processing unit (CPU).
15 . A method, comprising:
receiving a query with a relational database operation to perform on relational data, wherein the relational database operation is a group by operation on an input table of the relational data;
using a graphics pipeline to perform the relational database operation on the relational data; and
outputting a query result of the relational database operation, wherein the graphics pipeline performs:
a build phase on the relational data to create a hash table of the relational data, wherein the hash table maps the relational data to a stack of two-dimensional (2D) images and each 2D image of the stack of 2D images is created in response to a hash collision for the relational data; and
a probe phase that performs the relational database operation by rendering the relational database operation on the hash table and outputting the query result.
16 . The method of claim 15 , wherein the probe phase:
uses the stack of 2D images to map the relational data to positions defined by a grouping value;
generates pixels with the positions and the grouping values;
performs post processing that applies color to the pixels indicating an aggregate function and attribute values of the relational data; and
performs a blend function of the pixels based on the aggregation function and the query output includes the blended pixels.
17 . The method of claim 16 , wherein the aggregate function includes a count, a sum, a minimum, a maximum, or an average.
18 . The method of claim 15 , wherein the graphics pipeline includes a vertex shader, vertex post processing, a fragment shader, and fragment processing.
19 . The method of claim 15 , wherein the graphics pipeline is executed by a graphics processing unit (GPU) and the graphics pipeline includes a vertex shader, vertex post processing, a fragment shader, and fragment processing.